An abnormality detection and diagnosis method, system, device and medium for a central air conditioning water chiller

Abstract

The application provides a central air conditioner cold water unit anomaly detection and diagnosis method, system, device and medium, and belongs to the technical field of predictive maintenance of industrial equipment. The method steps are as follows: collecting and preprocessing time series data and log data of air conditioner unit operation; using the time series data of air conditioner unit operation to fine-tune and train a time series detection model, and using log data of air conditioner unit operation to fine-tune and train a large language model; inputting the preprocessed real-time operation data of the air conditioner unit into a time series prediction model for anomaly detection, and when an anomaly is detected, acquiring log data of the air conditioner unit operation in the abnormal period and inputting the log data into the large language model for anomaly diagnosis. The application can monitor the operation state of the cold water unit in real time, automatically trigger the diagnosis process when an anomaly is detected, realize the automation and intelligentization of operation and maintenance management, improve the anomaly response speed, and reduce the maintenance cost.

Description

Technical Field

[0001] This invention belongs to the field of predictive maintenance technology for industrial equipment, specifically relating to a method, system, equipment, and medium for detecting and diagnosing abnormalities in central air conditioning chiller units. Background Technology

[0002] Traditional central air conditioning chiller unit operation and maintenance practices have always relied on manual monitoring and diagnosis. This approach is not only inefficient, but also, due to limitations in professional knowledge and experience, it is often difficult to make quick and accurate judgments on complex abnormal patterns and deep-seated system problems. This not only increases maintenance costs, but also prolongs equipment downtime, affecting overall operating efficiency and production stability.

[0003] For anomaly detection, commonly used methods include deep learning-based approaches such as autoencoders, convolutional neural networks, and long short-term memory networks, which can learn and identify complex patterns and anomalous behaviors in data. While these methods have shown some effectiveness in image and speech recognition, they typically require large amounts of labeled data for training. This is often impractical in industrial applications because acquiring large amounts of labeled anomaly data is both time-consuming and expensive. Furthermore, deep learning models have poor interpretability; when a model predicts an anomaly, it is difficult to explain why such a prediction occurred. This is a significant drawback for industrial systems that require high reliability and interpretability.

[0004] Another way to detect anomalies is to analyze the operating logs of important air conditioning chillers. However, the mainstream methods of log analysis currently rely on regular expression matching, keyword search, and simple statistical analysis. However, these methods have obvious limitations, such as the inability to capture complex patterns and abnormal behaviors in log data. Furthermore, they require a lot of manual intervention to screen, classify, and interpret logs, which is inefficient and easily affected by human factors. Summary of the Invention

[0005] In a first aspect, embodiments of this application provide a method for detecting and diagnosing abnormalities in a central air conditioning chiller unit, comprising the following steps:

[0006] S1. Collect and preprocess the time-series data and log data of the air conditioning unit operation;

[0007] S2. Use the time series data of air conditioning unit operation to fine-tune the time series detection model, and use the log data of air conditioning unit operation to fine-tune the large language model;

[0008] S3. After preprocessing the real-time operation data of the air conditioning unit, input it into the time series prediction model for anomaly detection. When an anomaly is detected, obtain the log data of the air conditioning unit operation during the abnormal period and input it into the large language model for anomaly diagnosis.

[0009] Furthermore, the specific steps of step S1 are as follows:

[0010] S11. Collect time-series data of the air conditioning unit operation, including chilled water inlet and outlet temperatures, cooling water inlet and outlet temperatures, cooling water loop heat exchanger inlet and outlet temperatures, chilled water loop heat exchanger inlet and outlet temperatures, and compressor power.

[0011] S12. Clean the time-series data of the air conditioning unit operation by removing random noise through filtering algorithms, filling in missing data through linear interpolation, and unifying the data format and measurement units to complete standardization;

[0012] S13. Collect log data of the air conditioning unit, wherein the log data of the air conditioning unit includes equipment operating status, equipment alarm information, maintenance and diagnostic information, and historical anomaly information;

[0013] S14. Normalize the log data of the acquisition unit, truncate the log text according to the maximum input length of the large language model, and convert the log text into a format supported by the input of the large language model.

[0014] Furthermore, the specific steps of step S2 are as follows:

[0015] S21. Fine-tune the structure of the Moirai timing detection model based on the timing data of the air conditioning unit;

[0016] S22. Use the time series data of the air conditioning unit to train and optimize the fine-tuned Moirai time series detection model to obtain the optimized Moirai time series detection model.

[0017] S23. Fine-tune the backend layers of the Qwen large language model, add an adaptive layer to the feature extraction part, add an external knowledge base to the retrieval function of the RAG architecture, and limit the output anomaly report format to achieve matching with the anomaly diagnosis of air conditioning units.

[0018] Furthermore, the specific steps of step S21 are as follows:

[0019] S211. Divide the time-series data of the air conditioning unit operation into a training set and a validation set according to a set ratio;

[0020] S212. Select N patch sizes, and use multi-patch size input projection to segment the time series data of the training set and validation set according to each patch size into time windows to obtain N segmentation results;

[0021] S213. For each window of each segmentation result, use patch embedding to map it to a high-dimensional feature space to obtain the processed test set and validation set;

[0022] S214. Input the patched test set or validation set into the Moirai time series detection model, use the attention mechanism to process the data layer by layer, and adjust or add adaptive layers to the back-end layers of the Moirai time series detection model within a set range during the data processing to match the air conditioning unit anomaly detection task.

[0023] S215. Introduce a multi-layer arbitrary variable attention mechanism into the Moirai temporal detection model, and process sensor data of different quantities and types.

[0024] Furthermore, the specific steps of step S22 are as follows:

[0025] S221. Using the negative log-likelihood function of the predicted distribution of the normal operation data of the air conditioning unit as the objective function, the prediction error of the training samples is characterized.

[0026] S222. Using cross-validation, the Moirai time series detection model is iteratively optimized on the training and validation sets. Training is stopped when the objective function is minimized or the number of iterations is completed, resulting in the optimized Moirai time series detection model.

[0027] Furthermore, the specific steps of step S23 are as follows:

[0028] S231. Lock the front-end layers of the Qwen large language model, and adjust or add adaptive layers to the back-end layers within a set range.

[0029] S232. Add an adaptive layer to the feature extraction part of the Qwen large language model to adapt to the distribution characteristics of abnormal data of air conditioning units;

[0030] S233. Use the retrieval function in the RAG architecture to build an external knowledge base containing knowledge on the diagnosis of air conditioning unit anomalies, and introduce air conditioning chiller unit failure modes, handling suggestions and historical cases;

[0031] S234. Input the simulated air conditioning unit abnormality diagnosis related issues into the Qwen large language model, so that the Qwen large language model calls the relevant historical cases and diagnostic information in the search function of the RAG architecture, and adjusts the Qwen large language model according to the search results;

[0032] S235. Modify the output layer of the Qwen large language model, add formatted content to the output anomaly diagnosis report to match the annotation format of the air conditioning unit anomaly diagnosis; the formatted content includes the anomaly report header, anomaly description, root cause analysis and modification suggestions.

[0033] Furthermore, the specific steps of step S3 are as follows:

[0034] S31. Collect real-time operating data of the air conditioning unit, preprocess it, and then input it into the optimized Moirai time-series detection model;

[0035] S32. The Moirai time-series detection model analyzes various sensor data in the time and feature dimensions through an attention mechanism, compares the learned normal operation characteristics of the air conditioning unit with the input data, and obtains the prediction results of the input data.

[0036] S33. Analyze the prediction results, regard the probability of the data falling outside the normal range as the abnormal score, and determine whether the abnormal score exceeds the warning threshold.

[0037] If not, return to step S31;

[0038] If so, proceed to step S34;

[0039] S34. Trigger real-time alerts, identify abnormal time periods, and extract log data and time-series data of the air conditioning unit operation during the abnormal time periods;

[0040] S35. After preprocessing the log data of the air conditioning units during the extracted abnormal periods, input it along with the abnormal score into the fine-tuned Qwen large language model;

[0041] S36. The fine-tuned Qwen large language model uses the model's natural language processing capabilities to analyze log and sensor data during abnormal periods, and performs feature analysis on multimodal data. By comparing context and abnormal features, and combining historical data, it performs root cause analysis. At the same time, it activates the search function in RAG to retrieve abnormal records and diagnostic suggestions similar to the input data from the knowledge base.

[0042] S37. The fine-tuned Qwen large language model generates a structured anomaly diagnosis report through multimodal fusion, following the agreed format and combining anomaly scores, historical similar cases, and contextual information.

[0043] S38. Store the generated anomaly diagnosis reports in the knowledge base to build an air conditioning unit fault mode library, and further fine-tune the input and output of the Qwen large language model based on the anomaly diagnosis reports.

[0044] Secondly, embodiments of this application also provide a central air conditioning chiller unit anomaly detection and diagnosis system, including:

[0045] The data acquisition unit is used to collect time-series data and log data of the air conditioning unit operation and perform preprocessing.

[0046] The model fine-tuning unit is used to fine-tune the time-series detection model using time-series data of air conditioning unit operation, and to fine-tune the large language model using log data of air conditioning unit operation.

[0047] The anomaly detection and diagnosis unit is used to preprocess the real-time operation data of the air conditioning unit and input it into the time series prediction model for anomaly detection. When an anomaly is detected, the log data of the air conditioning unit operation during the abnormal period is obtained and input into the big data language model for anomaly diagnosis.

[0048] Thirdly, embodiments of this application also provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the central air conditioning chiller unit abnormality detection and diagnosis method as described in the first aspect.

[0049] Fourthly, embodiments of this application also provide a storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the central air conditioning chiller unit anomaly detection and diagnosis method as described in the first aspect.

[0050] As can be seen from the above technical solutions, the present invention has the following advantages:

[0051] The central air conditioning chiller unit anomaly detection and diagnosis method, system, equipment, and medium provided in this application collect and preprocess time-series and log data of the air conditioning unit, and use a finely tuned time-series detection model for anomaly detection. This enables accurate identification of anomaly patterns in the data, significantly improving detection accuracy. By leveraging efficient attention mechanisms and multi-layer arbitrary variable attention mechanisms, comprehensive analysis of multimodal data is achieved, enhancing detection efficiency. When an anomaly is detected, a finely tuned large language model is used to conduct in-depth analysis of the log data during the anomaly period. Combining historical data and similar cases in the knowledge base, the root cause of the anomaly can be intelligently diagnosed, and modification suggestions can be provided. This not only improves the intelligence level of diagnosis but also enhances the interpretability of diagnostic results, enabling maintenance personnel to more clearly understand the causes and solutions to the anomalies. The automated anomaly detection and diagnosis process reduces reliance on manual monitoring, minimizing misjudgments and omissions due to insufficient professional knowledge and experience. Simultaneously, timely anomaly warnings and accurate diagnostic results help maintenance personnel quickly locate and resolve problems, effectively shortening equipment downtime and reducing maintenance costs. Attached Figure Description

[0052] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0053] Figure 1 This is a flowchart illustrating an embodiment of the method for detecting and diagnosing abnormalities in a central air conditioning chiller unit according to the present invention.

[0054] Figure 2 This is a flowchart illustrating another embodiment of the method for detecting and diagnosing abnormalities in a central air conditioning chiller unit according to the present invention.

[0055] Figure 3 This is a schematic diagram of the central air conditioning chiller unit anomaly detection and diagnosis system of the present invention. Detailed Implementation

[0056] The various embodiments of this disclosure will be described more fully in the following detailed description of the methods for detecting and diagnosing abnormalities in central air conditioning chiller units. This disclosure may have various embodiments, and adjustments and changes may be made therein. However, it should be understood that there is no intention to limit the various embodiments of this disclosure to the specific embodiments disclosed herein, but rather this disclosure should be understood to cover all adjustments, equivalents, and / or alternatives falling within the spirit and scope of the various embodiments of this disclosure.

[0057] For example, in the traditional operation and maintenance model of central air conditioning chiller units, manual monitoring and inspection have always been dominant. However, this model is not only inefficient, but also limited by professional knowledge and experience, often making it difficult to make quick and accurate judgments when faced with complex anomalies and deep-seated system problems. This not only increases maintenance costs, but also prolongs equipment downtime, thereby affecting overall operational efficiency and production stability.

[0058] In the field of anomaly detection, the current popular approach is to employ deep learning-based techniques, such as autoencoders, convolutional neural networks, and long short-term memory networks. These techniques can learn and identify complex patterns and anomalous behaviors in data, and have shown some success in image and speech recognition. However, these methods face several challenges in practical industrial applications. On the one hand, they typically require large amounts of labeled data for training, and acquiring such labeled anomaly data in industrial environments is both time-consuming and costly. On the other hand, deep learning models have relatively poor interpretability; when a model predicts an anomaly, it is difficult to clearly explain the reason for the prediction. This is undoubtedly a significant drawback for industrial systems that require high reliability and interpretability.

[0059] Another common method for anomaly detection is analyzing the operating logs of critical air conditioning chillers. However, current mainstream log analysis methods still primarily rely on regular expression matching, keyword searching, and simple statistical analysis. The limitations of these methods are obvious: they cannot capture complex patterns and anomalous behaviors in log data. Furthermore, these methods require significant human intervention, including log screening, classification, and interpretation, which is not only inefficient but also susceptible to human error.

[0060] To address the aforementioned issues, this embodiment provides a method for anomaly detection and diagnosis of central air conditioning chiller units. This method can simultaneously and fully utilize time-series data and log data. The former enables early identification and warning of abnormal states, thereby reducing potential equipment malfunctions and downtime. The latter helps maintenance personnel quickly understand the causes of anomalies and provides targeted repair suggestions. Combining the two methods can significantly improve the operation and maintenance efficiency of chiller units, as well as significantly improve diagnostic accuracy and reduce equipment energy consumption, thereby enhancing both the operation and maintenance efficiency and diagnostic accuracy of chiller units.

[0061] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0062] Please see Figure 1 The diagram shows a flowchart of a method for detecting and diagnosing abnormalities in a central air conditioning chiller unit, according to a specific embodiment. The method includes the following steps:

[0063] S1. Collect and preprocess the time-series data and log data of the air conditioning unit operation;

[0064] It should be noted that the time-series data of the air conditioning unit operation is preprocessed to ensure that it matches the time-series detection model, and the log data of the air conditioning unit is preprocessed to ensure that it matches the large language model.

[0065] S2. Use the time series data of air conditioning unit operation to fine-tune the time series detection model, and use the log data of air conditioning unit operation to fine-tune the large language model;

[0066] It should be noted that fine-tuning the training time-series detection model will learn the multimodal data features of the normal operation of the air conditioning unit, and fine-tuning the training large language model will improve the ability to identify abnormal patterns of the air conditioning unit.

[0067] S3. After preprocessing the real-time operation data of the air conditioning unit, input it into the time series prediction model for anomaly detection. When an anomaly is detected, obtain the log data of the air conditioning unit operation during the abnormal period and input it into the big data language model for anomaly diagnosis.

[0068] It should be noted that by fusing time-series prediction models with large language models to detect and diagnose anomalies in air conditioning units, the operation and maintenance efficiency of air conditioning units can be greatly improved, diagnostic accuracy can be significantly increased, and energy consumption of equipment can be reduced.

[0069] The central air conditioning chiller unit anomaly detection and diagnosis method of this embodiment can monitor the operating status of the chiller unit in real time and automatically trigger the diagnosis process when an anomaly is detected, realizing the automation and intelligence of operation and maintenance management. It not only improves the anomaly response speed, but also reduces maintenance costs, which is of great significance for the stable operation of industrial chiller units.

[0070] Furthermore, as a refinement and extension of the specific implementation methods of the above embodiments, and to fully illustrate the specific implementation process in this embodiment, another method for detecting and diagnosing abnormalities in central air conditioning chiller units is provided, such as... Figure 2 As shown, the method includes the following steps:

[0071] S1. Collect and preprocess the time-series data and log data of the air conditioning unit operation; the specific steps of step S1 are as follows:

[0072] S11. Collect time-series data of the air conditioning unit operation, including chilled water inlet and outlet temperatures, cooling water inlet and outlet temperatures, cooling water loop heat exchanger inlet and outlet temperatures, chilled water loop heat exchanger inlet and outlet temperatures, and compressor power.

[0073] It should be noted that the inlet and outlet temperatures of chilled water, the inlet and outlet temperatures of cooling water, the inlet and outlet temperatures of the cooling water loop heat exchanger, and the near-outlet temperature of the chilled water loop heat exchanger are all measured using thermistors.

[0074] S12. Clean the time-series data of the air conditioning unit operation by removing random noise through filtering algorithms, filling in missing data through linear interpolation, and unifying the data format and measurement units to complete standardization;

[0075] It should be noted that filtering provides the signal-to-noise ratio of the data, filling in missing data ensures the integrity of the data, and unifying the data format and measurement units ensures that data from different sensor sources can be compared and processed under the same standard.

[0076] S13. Collect log data of the air conditioning unit, wherein the log data of the air conditioning unit includes equipment operating status, equipment alarm information, maintenance and diagnostic information, and historical anomaly information;

[0077] It should be noted that the equipment operating status data can include the on / off status changes of the chiller unit, and the status information of different working modes such as cooling, standby, and defrosting; the equipment alarm information can include high / low pressure alarms; overheating or overcooling alarms; overcurrent or undercurrent alarms; and sensor malfunction alarms; the maintenance and diagnostic information can include equipment maintenance records, including information on cleaning, lubrication, and inspection; and the historical anomaly information can include the time of occurrence, anomaly type, and handling result.

[0078] S14. Normalize the log data of the acquisition unit, truncate the log text according to the maximum input length of the large language model, and convert the log text into a format supported by the input of the large language model;

[0079] For example, the log text is truncated according to the maximum input length of the Qwen large language model, and the log text is converted into an input format supported by the Qwen large language model.

[0080] S2. Fine-tune the time-series detection model using time-series data from air conditioning unit operation, and fine-tune the large language model using log data from air conditioning unit operation; the specific steps of step S2 are as follows:

[0081] S21. Fine-tune the structure of the Moirai timing detection model based on the timing data of the air conditioning unit;

[0082] S22. Use the time series data of the air conditioning unit to train and optimize the fine-tuned Moirai time series detection model to obtain the optimized Moirai time series detection model.

[0083] It should be noted that the Moirai time series detection model was fine-tuned and trained using a normal time series dataset, allowing the Moirai time series detection model to learn the time series regularity information under the normal operation mode of the central air conditioning unit.

[0084] S23. Fine-tune the back-end layers of the Qwen large language model, add an adaptive layer to the feature extraction part, add an external knowledge base to the retrieval function of the RAG architecture, and limit the output anomaly report format to achieve matching with the anomaly diagnosis of air conditioning units.

[0085] It should be noted that the fine-tuning of the Qwen large language model is to prepare and process abnormal data. By adding the RAG mechanism, it can more accurately analyze the air conditioning unit log data and generate answers that better meet the requirements of abnormal diagnosis.

[0086] S3. After preprocessing the real-time operating data of the air conditioning unit, input it into the time series prediction model for anomaly detection. When an anomaly is detected, obtain the log data of the air conditioning unit's operation during the anomaly period and input it into the big data model for anomaly diagnosis. The specific steps of step S3 are as follows:

[0087] S31. Collect real-time operating data of the air conditioning unit, preprocess it, and then input it into the optimized Moirai time-series detection model;

[0088] The S32.Moirai time-series detection model analyzes various sensor data in the time and feature dimensions through an attention mechanism, compares the learned normal operation characteristics of the air conditioning unit with the input data, and obtains the prediction results of the input data.

[0089] It should be noted that the preprocessed data is input into the temporal Transformer architecture of the Moirai temporal detection model, and the attention mechanism is used to analyze the multimodal data in terms of time and feature dimensions. The Moirai temporal detection model compares the input data based on the normal operating characteristics of the equipment learned during pre-training and fine-tuning. Through the arbitrary variable attention mechanism, the model can adapt to different dimensions of chiller unit operating data and achieve comprehensive analysis of various sensor data. This process allows the Moirai temporal detection model to maintain accurate monitoring of each dimension when processing multiple sensor inputs.

[0090] S33. Analyze the prediction results, regard the probability of the data falling outside the normal range as the abnormal score, and determine whether the abnormal score exceeds the warning threshold.

[0091] If not, return to step S31;

[0092] If so, proceed to step S34;

[0093] It should be noted that the Moirai time series detection model generates predictions based on the mixed distribution probability of the input data. Data falling outside the normal distribution is considered an anomaly score to determine whether the current operating state deviates from the normal pattern. This assumes... It is input data collected in real time. The length of the time series. Indicates time Sensor data characteristics The distribution parameters learned by the model during the fine-tuning phase are represented by the following formula for calculating the anomaly score:

[0094]

[0095] S34. Trigger real-time alerts, identify abnormal time periods, and extract log data and time-series data of the air conditioning unit operation during the abnormal time periods;

[0096] It should be noted that if the abnormal score exceeds the warning threshold generated by the model, the system will trigger a real-time warning and automatically extract relevant log data for possible abnormal periods.

[0097] S35. After preprocessing the log data of the air conditioning units during the extracted abnormal periods, input it along with the abnormal score into the fine-tuned Qwen large language model;

[0098] It should be noted that all test results, including records of normal and abnormal states, are stored in the database. This historical data can be used for further analysis and model optimization, providing support for the improvement of long-term chiller unit operation status monitoring and maintenance strategies.

[0099] S36. The fine-tuned Qwen large language model uses the model's natural language processing capabilities to analyze log and sensor data during abnormal periods, and performs feature analysis on multimodal data. By comparing context and abnormal features, and combining historical data, it performs root cause analysis. At the same time, it activates the search function in RAG to retrieve abnormal records and diagnostic suggestions similar to the input data from the knowledge base.

[0100] It should be noted that multimodal data can include temperature, pressure, and alarm data, etc.

[0101] S37. The fine-tuned Qwen large language model generates a structured anomaly diagnosis report through multimodal fusion, following the agreed format and combining anomaly scores, historical similar cases, and contextual information.

[0102] It should be noted that during the model reasoning process, by comparing the context and abnormal features, the Qwen big language model identifies abnormal patterns, performs root cause analysis in combination with historical data, and simultaneously activates the search function in RAG to retrieve abnormal records and diagnostic suggestions similar to the input data from the knowledge base.

[0103] For example, let the feature vector of the current abnormal data be... The feature vector of the nth historical anomaly case in the knowledge base is The formula for calculating the similarity between the two is:

[0104]

[0105] By calculating the similarity between current abnormal data and historical abnormal cases, the system can identify the most similar abnormal patterns and add their root causes and solutions to the report:

[0106]

[0107] This ensures that the model can utilize historical data to improve the accuracy of root cause analysis during the diagnostic process;

[0108] S38. Store the generated anomaly diagnosis reports in the knowledge base to build an air conditioning unit fault mode library, and further fine-tune the input and output of the Qwen large language model based on the anomaly diagnosis reports;

[0109] To ensure the Qwen language model's output conforms to the specified report format, structured report templates are added to its input, such as report structure: anomaly description, root cause analysis, and remediation suggestions. Example reports and paragraph order guidance are provided to ensure the output report has a clear logical hierarchy and standardized structure, including anomaly description, root cause analysis, and remediation measures. Based on the inference results, the Qwen language model generates a detailed anomaly report, covering information such as the description of the anomaly event, the location of the root cause device, and the scope of impact. When generating the anomaly report, the Qwen language model combines anomaly scores, historical similar cases, and contextual information through multimodal fusion to generate a structured report. For example, consider time-series features... and log features The fusion vector is F. and F is a hyperparameter that adjusts the weights, used to balance the contributions of time-series and log features. The formula for calculating F is:

[0110]

[0111] Based on the fusion vector F, the Qwen large language model generates structured anomaly report content, including anomaly description, root cause analysis and remediation suggestions;

[0112] Furthermore, a feedback loop mechanism is established to allow maintenance personnel to provide feedback on generated reports, thereby optimizing the report generation quality of the Qwen big data language model. The generated diagnostic reports are stored in a knowledge base, gradually building and enriching the fault mode library for chiller units. In future diagnostics, when encountering similar anomalies, the Qwen big data language model can directly retrieve relevant records from the knowledge base, improving diagnostic efficiency. Based on feedback, the input format and prompts are gradually adjusted to optimize the Qwen big data language model's performance in generating anomaly reports that meet actual needs. Simultaneously, feedback will also be used to update the knowledge base, ensuring that the RAG module can access the latest and more adaptable anomaly handling cases during future inference processes.

[0113] In one embodiment of the present invention, based on steps S21, S22 and S23, a possible embodiment will be given below, and its specific implementation will be described in a non-limiting manner.

[0114] The specific steps of step S21 are as follows:

[0115] S211. Divide the time-series data of the air conditioning unit operation into a training set and a validation set according to a set ratio;

[0116] It should be noted that after the data preprocessing is completed, the processed air conditioning unit normal operation time series data is divided into training set and validation set, with a division ratio of 80%:20%, to ensure that the model can learn sufficient samples of normal equipment behavior during the training process.

[0117] S212. Select N patch sizes, and use multi-patch size input projection to segment the time series data of the training set and validation set according to each patch size into time windows to obtain N segmentation results;

[0118] It should be noted that a multi-patch size input projection operation is performed on the input time series data, dividing the time series data into time windows of specific lengths according to a variety of set patch sizes, so that the model can capture features at different time scales;

[0119] Set the input time series data of the chiller unit as a matrix ,in Indicates time The acquired feature vectors contain readings from multiple sensors. The multi-patch size input projection formula is as follows:

[0120]

[0121] Here, 𝑘 represents the size of the patch. By using different 𝑘 values, data can be segmented at different time scales; for example, small-sized patches are used to capture short-term fluctuations, while large-sized patches are used to capture long-term trends.

[0122] Two patch sizes are selected as an example: k 1=3 and k 2=10

[0123] for k 1=3, the generated time window is: [ x 1, x 4, x 7,…, x 997 (A total of 333 windows)

[0124] for k 2=10, the generated time window is: [ x 1, x 11 , x 21 ,…, x 991 (Total 99 windows)

[0125] S213. For each window of each segmentation result, use patch embedding to map it to a high-dimensional feature space to obtain the processed test set and validation set;

[0126] It should be noted that patch embedding is used to map these windows into a high-dimensional feature space so that the model can effectively capture the temporal patterns and correlations of device operation data in subsequent layers.

[0127] S214. Input the patched test set or validation set into the Moirai time series detection model, use the attention mechanism to process the data layer by layer, and adjust or add adaptive layers to the back-end layers of the Moirai time series detection model within a set range during the data processing to match the air conditioning unit anomaly detection task.

[0128] S215. Introduce a multi-layer arbitrary variable attention mechanism into the Moirai temporal detection model, and process sensor data of different quantities and types;

[0129] It should be noted that the embedded time-series data is input into the core temporal Transformer architecture of the Moirai time-series detection model. The model utilizes its attention mechanism to process the data layer by layer. During the fine-tuning stage, some pre-trained parameters are locked, and only the last few layers or added adaptive layers are fine-tuned to adapt to the specific data distribution of the chiller unit. A multi-layer arbitrary variable attention mechanism is adopted. This mechanism allows the model to maintain the independence between dimensions when processing different amounts and types of sensor data, thereby improving the adaptability of the Moirai time-series detection model to multimodal data of chiller units.

[0130] The specific steps of step S22 are as follows:

[0131] S221. Using the negative log-likelihood function of the predicted distribution of the normal operation data of the air conditioning unit as the objective function, the prediction error of the training samples is characterized.

[0132] It should be noted that during the fine-tuning process, a specific objective function is used to optimize the parameters of the Moirai time series detection model; this objective function is usually the negative log-likelihood, ensuring that the Moirai time series detection model is sensitive to anomalous patterns while predicting normal data distributions.

[0133] Assume the model output is the predicted distribution of normal data. ,in This indicates the normal operating data of the chiller unit. The number of training samples is given by the objective function:

[0134]

[0135] S222. Using cross-validation, the Moirai time series detection model is iteratively optimized on the training and validation sets. Training is stopped when the objective function is minimized or the number of iterations is completed, resulting in the optimized Moirai time series detection model.

[0136] It should be noted that cross-validation is used to iteratively optimize on the training and validation sets to reduce overfitting and make the Moirai time series detection model more generalizable.

[0137] The specific steps of step S23 are as follows:

[0138] S231. Lock the front-end layers of the Qwen large language model, and adjust or add adaptive layers to the back-end layers within a set range.

[0139] It should be noted that, in order to preserve the basic language understanding ability acquired by the Qwen large language model during pre-training, the parameters of the first few layers will be locked during fine-tuning, and only the parameters of the last few layers will be fine-tuned or adaptive layers will be added, so that the Qwen large language model can focus on the abnormal characteristics of the chiller unit based on the original general knowledge.

[0140] S232. Add an adaptive layer to the feature extraction part of the Qwen large language model to adapt to the distribution characteristics of abnormal data of air conditioning units;

[0141] It should be noted that task-specific adaptive layers, such as small-scale fully connected layers or convolutional layers, will be added to the feature extraction module of the Qwen big language model. These layers are specifically designed to adapt to the distribution characteristics of abnormal data from chiller units, so that the Qwen big language model can better capture the subtle differences in abnormal states of chiller units. By learning specific abnormal patterns of chiller units, the adaptive layers make the Qwen big language model more sensitive in distinguishing between normal and abnormal states.

[0142] S233. Use the retrieval function in the RAG architecture to build an external knowledge base containing knowledge on the diagnosis of air conditioning unit anomalies, and introduce air conditioning chiller unit failure modes, handling suggestions and historical cases;

[0143] S234. Input the simulated air conditioning unit abnormality diagnosis related issues into the Qwen large language model, so that the Qwen large language model calls the relevant historical cases and diagnostic information in the search function of the RAG architecture, and adjusts the Qwen large language model according to the search results;

[0144] It should be noted that by using an external knowledge base, the ability of Qwen's large language model to identify abnormal patterns in chiller units can be improved, while enriching its diagnostic output.

[0145] S235. Modify the output layer of the Qwen large language model, add formatted content to the output anomaly diagnosis report to match the annotation format of the air conditioning unit anomaly diagnosis; the formatted content includes the anomaly report header, anomaly description, root cause analysis and modification suggestions;

[0146] It should be noted that the output layer of the model has been modified to make it more suitable for generating structured anomaly report content, ensuring that the Qwen large language model can cover the key anomaly information of the chiller unit when generating reports and can conform to the standardized structure of diagnostic reports.

[0147] This invention can simultaneously and fully mine time-series data and log data to achieve early identification and warning of abnormal states, thereby reducing potential equipment anomalies and downtime, helping maintenance personnel to quickly understand the causes of anomalies, and providing targeted repair suggestions. Combining the two can significantly improve the operation and maintenance efficiency of chiller units, as well as significantly improve diagnostic accuracy and reduce equipment energy consumption.

[0148] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0149] like Figure 3 As shown, the following are embodiments of the central air conditioning chiller unit anomaly detection and diagnosis system provided in this disclosure. This system and the central air conditioning chiller unit anomaly detection and diagnosis methods in the above embodiments belong to the same inventive concept. For details not described in detail in the embodiments of the central air conditioning chiller unit anomaly detection and diagnosis system, please refer to the embodiments of the above central air conditioning chiller unit anomaly detection and diagnosis methods.

[0150] The system includes:

[0151] The data acquisition unit is used to collect time-series data and log data of the air conditioning unit operation and perform preprocessing.

[0152] The model fine-tuning unit is used to fine-tune the time-series detection model using time-series data of air conditioning unit operation, and to fine-tune the large language model using log data of air conditioning unit operation.

[0153] The anomaly detection and diagnosis unit is used to preprocess the real-time operation data of the air conditioning unit and input it into the time series prediction model for anomaly detection. When an anomaly is detected, the log data of the air conditioning unit operation during the abnormal period is obtained and input into the big data language model for anomaly diagnosis.

[0154] The central air conditioning chiller unit anomaly detection and diagnosis system in this embodiment can monitor the operating status of the chiller unit in real time and automatically trigger the diagnosis process when an anomaly is detected. It realizes the automation and intelligence of operation and maintenance management, which not only improves the anomaly response speed, but also reduces maintenance costs. It is of great significance for the stable operation of industrial chiller units.

[0155] The method for detecting and diagnosing abnormalities in central air conditioning chiller units provided in this application can be applied to electronic devices. Those skilled in the art will understand that the electronic device structure involved in the embodiments of this invention does not constitute a limitation on the electronic device. An electronic device may include more or fewer components than illustrated, or combine certain components, or have different component arrangements. In the embodiments of this invention, the electronic device includes, but is not limited to, laptop computers, desktop computers, workbenches, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the embodiments of this application described and / or claimed herein.

[0156] Electronic devices may include processors, external memory interfaces, internal memory, universal serial bus (USB) interfaces, charging management modules, power management modules, batteries, wireless communication modules, audio modules, speakers, microphones, sensor modules, buttons, cameras, displays, and SIM card interfaces, etc.

[0157] It is understood that the structures illustrated in the embodiments of this application do not constitute a specific limitation on the electronic device. In other embodiments of this application, the electronic device may include more or fewer components than illustrated, or combine some components, or split some components, or have different component arrangements. The illustrated components may be implemented in hardware, software, or a combination of software and hardware.

[0158] A processor may include one or more processing units, such as: a central processing unit (CPU), an application processor (AP), a modem processor, a graphics processing unit (GPU), an image signal processor (ISP), a controller, memory, a video codec, a digital signal processor (DSP), a baseband processor, and / or a neural network processing unit (NPU). Different processing units may be independent devices or integrated into one or more processors.

[0159] The processor can serve as the nerve center and command center of an electronic device. The controller can generate operation control signals based on the instruction opcode and timing signals to control the fetching and execution of instructions.

[0160] The processor may also include memory for storing instructions and data. In some embodiments, the memory in the processor is a cache memory. This memory can store instructions or data that the processor has just used or that are used repeatedly. If the processor needs to use the instruction or data again, it can retrieve it directly from this memory. This avoids repeated accesses, reduces processor latency, and thus improves system efficiency.

[0161] The aforementioned electronic equipment implements the technical solution of the central air conditioning chiller unit anomaly detection and diagnosis method of this application: collecting and preprocessing time-series data and log data of the air conditioning unit operation; using the time-series data of the air conditioning unit operation to fine-tune and train the time-series detection model, and using the log data of the air conditioning unit operation to fine-tune and train the large language model; inputting the preprocessed real-time operation data of the air conditioning unit into the time-series prediction model for anomaly detection; and when an anomaly is detected, obtaining the log data of the air conditioning unit operation during the anomaly period and inputting it into the large language model for anomaly diagnosis. This achieves real-time monitoring of the chiller unit's operating status and automatically triggers the diagnostic process when an anomaly is detected, realizing the automation and intelligence of operation and maintenance management. It not only improves the anomaly response speed but also reduces maintenance costs.

[0162] The storage medium provided in this application stores a program product capable of implementing a method for detecting and diagnosing abnormalities in a central air conditioning chiller unit.

[0163] The method for anomaly detection and diagnosis of central air conditioning chiller units includes: collecting and preprocessing time-series and log data of the air conditioning unit operation; fine-tuning and training the time-series detection model using the time-series data of the air conditioning unit operation, and fine-tuning and training the large language model using the log data of the air conditioning unit operation; inputting the preprocessed real-time operation data of the air conditioning unit into the time-series prediction model for anomaly detection; when an anomaly is detected, obtaining the log data of the air conditioning unit operation during the anomaly period and inputting it into the large language model for anomaly diagnosis.

[0164] In some possible implementations, the central air conditioning chiller unit anomaly detection and diagnosis method of this disclosure can be implemented as a program product, which includes program code. When the program product is run on a terminal device, the program code is used to cause the terminal device to perform the steps described in the "Exemplary Methods" section above according to various exemplary embodiments of this disclosure.

[0165] The storage medium disclosed herein may be any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of readable storage media include: an electrical connection having one or more wires, a portable disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof.

[0166] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for detecting and diagnosing abnormalities in a central air conditioning chiller unit, characterized in that, Includes the following steps: S1. Collect and preprocess the time-series data and log data of the air conditioning unit operation; S2. Fine-tune the time-series detection model using time-series data from air conditioning unit operation, and fine-tune the large language model using log data from air conditioning unit operation; the specific steps of step S2 are as follows: S21. Fine-tune the structure of the Moirai timing detection model based on the timing data of the air conditioning unit; S22. Use the time series data of the air conditioning unit to train and optimize the fine-tuned Moirai time series detection model to obtain the optimized Moirai time series detection model. S23. Fine-tune the back-end layers of the Qwen large language model, add an adaptive layer to the feature extraction part, add an external knowledge base to the retrieval function of the RAG architecture, and limit the output anomaly report format to achieve matching with the anomaly diagnosis of air conditioning units. S3. After preprocessing the real-time operating data of the air conditioning unit, input it into the time series prediction model for anomaly detection. When an anomaly is detected, obtain the log data of the air conditioning unit's operation during the anomaly period and input it into the large language model for anomaly diagnosis. The specific steps of step S3 are as follows: S31. Collect real-time operating data of the air conditioning unit, preprocess it, and then input it into the optimized Moirai time-series detection model; S32. The Moirai time-series detection model analyzes various sensor data in the time and feature dimensions through an attention mechanism, compares the learned normal operation characteristics of the air conditioning unit with the input data, and obtains the prediction results of the input data. S33. Analyze the prediction results, regard the probability of the data falling outside the normal range as the abnormal score, and determine whether the abnormal score exceeds the warning threshold. If not, return to step S31; If so, proceed to step S34; S34. Trigger real-time alerts, identify abnormal time periods, and extract log data and time-series data of the air conditioning unit operation during the abnormal time periods; S35. After preprocessing the log data of the air conditioning units during the extracted abnormal periods, input it along with the abnormal score into the fine-tuned Qwen large language model; S36. The fine-tuned Qwen large language model uses the model's natural language processing capabilities to analyze log and sensor data during abnormal periods, and performs feature analysis on multimodal data. By comparing context and abnormal features, and combining historical data, it performs root cause analysis. At the same time, it activates the search function in RAG to retrieve abnormal records and diagnostic suggestions similar to the input data from the knowledge base. S37. The fine-tuned Qwen large language model generates a structured anomaly diagnosis report through multimodal fusion, following the agreed format and combining anomaly scores, historical similar cases, and contextual information. S38. Store the generated anomaly diagnosis reports in the knowledge base to build an air conditioning unit fault mode library, and further fine-tune the input and output of the Qwen large language model based on the anomaly diagnosis reports.

2. The method for detecting and diagnosing abnormalities in a central air conditioning chiller unit as described in claim 1, characterized in that, The specific steps of step S1 are as follows: S11. Collect time-series data of the air conditioning unit operation, including chilled water inlet and outlet temperatures, cooling water inlet and outlet temperatures, cooling water loop heat exchanger inlet and outlet temperatures, chilled water loop heat exchanger inlet and outlet temperatures, and compressor power. S12. Clean the time-series data of the air conditioning unit operation by removing random noise through filtering algorithms, filling in missing data through linear interpolation, and unifying the data format and measurement units to complete standardization; S13. Collect log data of the air conditioning unit, wherein the log data of the air conditioning unit includes equipment operating status, equipment alarm information, maintenance and diagnostic information, and historical anomaly information; S14. Normalize the log data of the acquisition unit, truncate the log text according to the maximum input length of the large language model, and convert the log text into a format supported by the input of the large language model.

3. The method for abnormal detection and diagnosis of central air conditioning chiller units as described in claim 2, characterized in that, The specific steps of step S21 are as follows: S211. Divide the time-series data of the air conditioning unit operation into a training set and a validation set according to a set ratio; S212. Select N patch sizes, and use multi-patch size input projection to segment the time series data of the training set and validation set according to each patch size into time windows to obtain N segmentation results; S213. For each window of each segmentation result, use patch embedding to map it to a high-dimensional feature space to obtain the processed test set and validation set; S214. Input the patched test set or validation set into the Moirai time series detection model, use the attention mechanism to process the data layer by layer, and adjust or add adaptive layers to the back-end layers of the Moirai time series detection model within a set range during the data processing to match the air conditioning unit anomaly detection task. S215. Introduce a multi-layer arbitrary variable attention mechanism into the Moirai temporal detection model, and process sensor data of different quantities and types.

4. The method for detecting and diagnosing abnormalities in a central air conditioning chiller unit as described in claim 2, characterized in that, The specific steps of step S22 are as follows: S221. Using the negative log-likelihood function of the predicted distribution of the normal operation data of the air conditioning unit as the objective function, the prediction error of the training samples is characterized. S222. Using cross-validation, the Moirai time series detection model is iteratively optimized on the training and validation sets. Training is stopped when the objective function is minimized or the number of iterations is completed, resulting in the optimized Moirai time series detection model.

5. The method for detecting and diagnosing abnormalities in a central air conditioning chiller unit as described in claim 2, characterized in that, The specific steps of step S23 are as follows: S231. Lock the front-end layers of the Qwen large language model, and adjust or add adaptive layers to the back-end layers within a set range. S232. Add an adaptive layer to the feature extraction part of the Qwen large language model to adapt to the distribution characteristics of abnormal data of air conditioning units; S233. Use the retrieval function in the RAG architecture to build an external knowledge base containing knowledge on the diagnosis of air conditioning unit anomalies, and introduce air conditioning chiller unit failure modes, handling suggestions and historical cases; S234. Input the simulated air conditioning unit abnormality diagnosis related issues into the Qwen large language model, so that the Qwen large language model calls the relevant historical cases and diagnostic information in the search function of the RAG architecture, and adjusts the Qwen large language model according to the search results; S235. Modify the output layer of the Qwen large language model, add formatted content to the output anomaly diagnosis report to match the annotation format of the air conditioning unit anomaly diagnosis; the formatted content includes the anomaly report header, anomaly description, root cause analysis and modification suggestions.

6. A system for detecting and diagnosing abnormalities in a central air conditioning chiller unit, employing the method for detecting and diagnosing abnormalities in a central air conditioning chiller unit as described in any one of claims 1-5, characterized in that, include: The data acquisition unit is used to collect time-series data and log data of the air conditioning unit operation and perform preprocessing. The model fine-tuning unit is used to fine-tune the time-series detection model using time-series data of air conditioning unit operation, and to fine-tune the large language model using log data of air conditioning unit operation. The anomaly detection and diagnosis unit is used to preprocess the real-time operation data of the air conditioning unit and input it into the time series prediction model for anomaly detection. When an anomaly is detected, the log data of the air conditioning unit operation during the abnormal period is obtained and input into the big data language model for anomaly diagnosis.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the central air conditioning chiller unit abnormality detection and diagnosis method as described in any one of claims 1 to 5.

8. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the method for abnormal detection and diagnosis of central air conditioning chiller units as described in any one of claims 1 to 5.

Images